1) Commit to long term API support. It makes no sense to want a diverse/sustained app ecosystem, then throw away existing apps because the API changed and nobody is there to fix it. I can run win95 apps. I can compile command line unix apps older than that. But gtk/gnome for more than a few years ago gets hard.
2) Decide on minimal supported target hardware. Gnome seems to put in a lot of polish for top end hardware. Meanwhile, the average piece of hardware is a cheap computer. If it runs Linux, it's possibly a hand me down. Gnome should run well on a 10 year old craptop with low-res screen.
3) Decide on a user model. Gnome seems focused on the average 90-year blind computer-illiterate Linux enthousiasts. I'd like to see them focus on 1 or more persona.